Luca Gravina

Indeed `reinit!` serves the purpose but it has one problem that might be an interesting issue. Consider the following simple problem: ``` u0 = 1 tl = LinRange(0,1,1001) tspan =...

I tried solving by using ``` function affect_1!(integrator) integrator.u = sol.u[select(t0, tl)] integrator.t = t0 reinit!(integrator, integrator.u; t0=t0, erase_sol=false) idx = findlast(integrator.sol.t .

This is perfect. Indeed it works. The only problem now is that if one has a PresetTimeCallback this interferes with the reset of the integrator. Specifically, if I take again...

Regarding the mean of the gradient, I would point out that in the example above you are only averaging the Jacobian and therefore what you see is not the gradient,...

I do not see a problem with how it is now, but if it is unclear maybe something could be added to the doc?

I have personally used both versions before. Maybe having two flags implementing both variants and returning an error if both are true could be a solution that would then make...

I really like the new implementation, its basically the same as we have here and from rough tests it performs the same too. One difference I note with the code...